Underwater moving target processing method and device
By preprocessing and clustering operations of initial full Ping point cloud data of water bottom, the problem of difficult separation of three-dimensional bottom dynamic targets is solved, more accurate bottom target detection is achieved, the calculation complexity is reduced, and a structure is provided data format.
Patent Information
- Application Number
- CN202411050388.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2024-12-03
AI Technical Summary
The prior art lacks effective inter-frame difference technology for background and dynamic target preprocessing in three-dimensional point cloud data, especially in the field of three-dimensional imaging sonar, and data acquisition difficulties lead to fewer algorithms.
By obtaining the initial full Ping point cloud data at the bottom, preprocessing is performed to reduce dimensionality, clustering operations are used to identify the coordinates of the category center point, and combining the clustering results to distinguish background and dynamic targets, using voxelized downsampling and projection to reduce dimensionality, DBSCAN is used The algorithm performs clustering analysis.
It improves the accuracy of bottom target detection, reduces the computational complexity, solves the problem of difficulty in separation of three-dimensional bottom dynamic targets, and provides a structured data format for easy identification and analysis.
Smart Images

Figure CN119068324A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of underwater moving targets, and in particular to an underwater moving target processing method and device. Background Art
[0002] In two-dimensional visual images, the inter-frame difference method is often used to preprocess the background, noise and moving targets, and further to detect moving targets. By taking the difference between the pixel points at each position on adjacent frame images and comparing the absolute value of the difference with the preset threshold value, the detection and extraction function of moving targets between adjacent frames can be realized. Based on this, in the preprocessing of the background and moving targets of two-dimensional images, methods such as the three-frame difference method, background attenuation method and optical flow field have emerged. In the field of three-dimensional point cloud images, especially in the preprocessing of moving targets and the extraction of background and targets in three-dimensional imaging sonar, there are relatively few public methods and technologies.
[0003] The existing inter-frame difference moving target detection method is mainly applicable to two-dimensional images or videos. For the background and moving target preprocessing in three-dimensional point cloud data, due to the differences in the number of point clouds and spatial positions in adjacent frames, effective inter-frame difference technology has not yet been made public.
[0004] In addition, in the detection preprocessing method for three-dimensional imaging sonar moving targets, the frame is called Ping. Due to the limitations of industry data sets (the difficulty of sonar signal data collection, high collection cost, and scarce data), there are few related algorithms disclosed in this field. Summary of the invention
[0005] The present invention provides a method and device for processing underwater moving targets, so as to solve the problem of great difficulty in separating three-dimensional underwater moving targets in the prior art.
[0006] In a first aspect, the present application provides a method for processing underwater moving targets, comprising:
[0007] Get the initial full Ping point cloud data of the bottom;
[0008] Preprocess the initial full Ping point cloud data of the bottom to obtain the reduced-dimensional full Ping point cloud data;
[0009] Performing a clustering operation on the dimension-reduced full Ping point cloud data to obtain the coordinates of the cluster center points of each category in the single Ping point cloud data;
[0010] Perform clustering operation on the coordinates of cluster center points of each category in single Ping point cloud data to obtain clustering results of each category of the whole Ping point cloud data;
[0011] According to the clustering results of each category of the full Ping point cloud data, the first underwater background and the first underwater moving target in the initial full Ping point cloud data are determined.
[0012] This application first achieves data dimensionality reduction through a preprocessing step, converting complex three-dimensional point cloud data into a two-dimensional form that is easier to process, thereby reducing the complexity of the calculation and retaining key spatial information. Then, a clustering operation is performed on the full Ping point cloud data after dimensionality reduction, effectively dividing the data into different categories, laying the foundation for subsequent analysis. The center point coordinates of each category are further calculated to provide a representative location reference for the point cloud data within each single Ping. By clustering these center point coordinates again, this method can identify the spatial distribution characteristics of the underwater background and moving targets on the scale of the full Ping data. Finally, combined with the full Ping clustering results, this application can distinguish and determine the first underwater background and the first underwater moving target in the initial full Ping point cloud data of the underwater. This process not only improves the accuracy of underwater target detection, but also solves the problem of difficulty in separating three-dimensional underwater moving targets in the prior art.
[0013] As a preferred embodiment of the first aspect, the preprocessing of the initial full Ping point cloud data of the bottom to obtain the reduced-dimensional full Ping point cloud data is specifically as follows:
[0014] Get the initial full ping point data of the bottom;
[0015] Perform voxel downsampling operation on the initial full Ping point cloud data to obtain downsampled full Ping point cloud data;
[0016] The downsampled full Ping point cloud data is projected in the XZ direction to obtain the reduced-dimensional full Ping point cloud data.
[0017] In this preferred embodiment, the present application first obtains the initial full Ping point cloud data of the bottom of the water, and effectively reduces the amount of data through voxelized downsampling operations. This step not only reduces the complexity of the data, but also helps to remove noise and retain key spatial features. Subsequently, the downsampled point cloud data is projected in the XZ direction to further convert the three-dimensional data into a two-dimensional representation. This dimensionality reduction process not only simplifies the data processing requirements, but also makes the visualization and subsequent analysis of the data more intuitive and efficient. This preprocessing process provides a structured and easy-to-operate format for the point cloud data of the underwater environment, laying a solid foundation for accurately identifying and analyzing the underwater background and moving targets.
[0018] As a preferred embodiment of the first aspect, the clustering operation is performed on the dimension-reduced full Ping point cloud data to obtain the coordinates of the cluster center points of each category in the single Ping point cloud data, specifically:
[0019] Performing a clustering operation on the dimension-reduced full Ping point cloud data to obtain clustering results of each category in the single Ping point cloud data;
[0020] Calculate the three-dimensional center point coordinates of the clustering results of each category in the single Ping point cloud data respectively;
[0021] The three-dimensional center point coordinates are uniformly displayed in one coordinate system to obtain a first coordinate system;
[0022] The coordinates of all Ping center points are clustered in the first coordinate system to obtain the coordinates of cluster center points of each category in the single Ping point cloud data.
[0023] In this preferred embodiment, the present application can accurately locate the central tendency of each category in three-dimensional space by calculating the center point coordinates of the clustering results of each category in a single Ping point cloud data. First, obtain the various categories obtained by cluster analysis in the single Ping point cloud data, which represent the distribution areas of different targets or backgrounds in the data. Then, further calculate the three-dimensional hollow point coordinates of these categories, which provide the geometric center of the category space distribution. Finally, select and retain the most representative center point coordinates from these hollow point coordinates. This step not only simplifies the data, but also enhances the description of the core position of each category.
[0024] As a preferred embodiment of the first aspect, after performing a clustering operation on the dimension-reduced full Ping point cloud data, it further includes:
[0025] Get the clustering results of each category in a single Ping point cloud data;
[0026] Calculate the intensity variance value of the clustering results of each category in the single Ping point cloud data;
[0027] Compare the intensity variance values of the clustering results of each category in the single Ping point cloud data;
[0028] According to the intensity variance value of the clustering results of each category in the single Ping point cloud data, the second underwater background and the second underwater moving target are separated, and the second underwater background is used to assist in judging the first underwater background; among them, the second underwater background has the largest intensity variance value.
[0029] As a preferred embodiment of the first aspect, the second underwater background is used to assist in determining the first underwater background, specifically:
[0030] The second underwater background is used as a strongly correlated confidence indicator for distinguishing the first underwater background.
[0031] In this preferred embodiment, the present application can identify different categories in a single Ping point cloud data by first performing a clustering operation on the dimensionally reduced full Ping point cloud data, which lays the foundation for further analysis. Subsequently, the intensity variance value of each clustering result is calculated. This step reveals the distribution consistency of the point cloud intensity within each category. By comparing the sizes of these variance values, it can be inferred that the cluster with a larger variance value may represent the underwater background, because the background area usually contains a variety of different reflection sources, while the cluster with a smaller variance value may represent a moving target in the water, because the reflection characteristics of the moving target are relatively consistent. Based on this reasoning, the cluster with the largest intensity variance value is identified as the second underwater background, and the cluster with the smallest intensity variance value is identified as the second underwater moving target. This method not only enhances the understanding of the underwater environment, but also provides an auxiliary means to verify and assist in judging the classification results of the underwater background and moving targets previously obtained by full Ping clustering, thereby improving the accuracy and reliability of the overall recognition process.
[0032] As a preferred embodiment of the first aspect, clustering operation is performed on the center point coordinates of each category in the single Ping point cloud data to obtain clustering results of each category of the full Ping point cloud data; according to the clustering results of each category of the full Ping point cloud data, the first underwater background and the first underwater moving target in the underwater initial full Ping point cloud data are determined, specifically:
[0033] Get the center point coordinates of each category in a single Ping point cloud data;
[0034] Display the center point coordinates of each category in all single Ping point cloud data in the same coordinate system to obtain a second coordinate system;
[0035] Initialize the domain radius value and domain point value of the DBSCAN clustering algorithm;
[0036] Using the DBSCAN clustering algorithm to cluster the center point coordinates of each category in the single Ping point cloud data in the second coordinate system, to obtain clustering results of each category of the full Ping point cloud data;
[0037] According to the clustering results of each category of the full Ping point cloud data, the center point coordinate change trend of the clustering results of each category of the full Ping point cloud data is analyzed, and the first underwater background and the first underwater moving target are separated according to the center point coordinate change trend; among them, the first underwater background has the smallest center point coordinate change trend, and the first underwater moving target has the largest center point coordinate change trend.
[0038] In this preferred embodiment, the present application effectively integrates the information in the full Ping point cloud data by accurately clustering the center point coordinates of each category in the single Ping point cloud data, thereby identifying the underwater background and underwater moving targets. First, the center point coordinates of each category in the single Ping are obtained, and these coordinates represent the spatial concentration trend of each cluster. Then, by initializing and applying the DBSCAN clustering algorithm, these center point coordinates are clustered within the full Ping data range, and the clustering results of each category of the full Ping point cloud data are obtained. Then, the change trend of the center point coordinates in these clustering results is analyzed. This step reveals the dynamic characteristics of the point cloud data in the time series. Finally, according to the change trend of the center point coordinates, the cluster with the smallest change trend is identified as the first underwater background, and the cluster with the largest change trend is identified as the first underwater moving target. This method utilizes the spatial and temporal information of the point cloud data, and through clustering and trend analysis, it improves the recognition accuracy of underwater static background and dynamic targets, and provides a more accurate and reliable technical means for monitoring and analyzing underwater environments.
[0039] In a second aspect, the present application provides an underwater moving target processing device. The underwater moving target processing device includes an acquisition module, a preprocessing module and a calculation module;
[0040] The acquisition module is used to obtain the initial full Ping point cloud data of the bottom;
[0041] The preprocessing module is used to preprocess the initial full Ping point cloud data of the bottom to obtain the reduced-dimensional full Ping point cloud data;
[0042] The computing module performs a clustering operation on the dimension-reduced full Ping point cloud data to obtain clustering results of each category in the single Ping point cloud data;
[0043] Calculate the center point coordinates of the clustering results of each category in the single Ping point cloud data to obtain the center point coordinates of each category in the single Ping point cloud data;
[0044] Perform clustering operation on the center point coordinates of each category in the single Ping point cloud data to obtain the clustering results of each category of the full Ping point cloud data;
[0045] According to the clustering results of each category of the full Ping point cloud data, the first underwater background and the first underwater moving target in the initial full Ping point cloud data are determined.
[0046] This application uses three modules to divide the work and coordinate the work to better complete the separation of underwater moving targets and underwater backgrounds. This application first achieves data dimensionality reduction through a preprocessing step, converting complex three-dimensional point cloud data into a two-dimensional form that is easier to process, thereby reducing the complexity of the calculation and retaining key spatial information. Then, a clustering operation is performed on the full Ping point cloud data after dimensionality reduction to effectively divide the data into different categories, laying the foundation for subsequent analysis. The center point coordinates of each category are further calculated to provide a representative location reference for the point cloud data in each single Ping. By clustering these center point coordinates again, this application can identify the spatial distribution characteristics of the underwater background and moving targets on the scale of the full Ping data. Finally, combined with the full Ping clustering results, this method can distinguish and determine the first underwater background and the first underwater moving target in the initial full Ping point cloud data of the bottom. This process not only improves the accuracy of underwater target detection, but also solves the problem of difficulty in separating three-dimensional underwater moving targets in the prior art.
[0047] As a preferred embodiment of the second aspect, the preprocessing of the initial full Ping point cloud data of the bottom to obtain the reduced-dimensional full Ping point cloud data is specifically as follows:
[0048] Get the initial full ping point data of the bottom;
[0049] Perform voxel downsampling operation on the initial full Ping point cloud data to obtain downsampled full Ping point cloud data;
[0050] The downsampled full Ping point cloud data is projected in the XZ direction to obtain the reduced-dimensional full Ping point cloud data.
[0051] In this preferred embodiment, the present application first obtains the initial full Ping point cloud data of the bottom of the water, and effectively reduces the amount of data through voxelized downsampling operations. This step not only reduces the complexity of the data, but also helps to remove noise and retain key spatial features. Subsequently, the downsampled point cloud data is projected in the XZ direction to further convert the three-dimensional data into a two-dimensional representation. This dimensionality reduction process not only simplifies the data processing requirements, but also makes the visualization and subsequent analysis of the data more intuitive and efficient. This preprocessing process provides a structured and easy-to-operate format for the point cloud data of the underwater environment, laying a solid foundation for accurately identifying and analyzing the underwater background and moving targets.
[0052] As a preferred embodiment of the second aspect, the clustering operation is performed on the dimension-reduced full Ping point cloud data to obtain the coordinates of the cluster center points of each category in the single Ping point cloud data, specifically:
[0053] Performing a clustering operation on the dimension-reduced full Ping point cloud data to obtain clustering results of each category in the single Ping point cloud data;
[0054] Calculate the three-dimensional center point coordinates of the clustering results of each category in the single Ping point cloud data respectively;
[0055] The three-dimensional center point coordinates are uniformly displayed in one coordinate system to obtain a first coordinate system;
[0056] The coordinates of all Ping center points are clustered in the first coordinate system to obtain the coordinates of cluster center points of each category in the single Ping point cloud data.
[0057] In this preferred embodiment, the present application can accurately locate the central tendency of each category in three-dimensional space by calculating the center point coordinates of the clustering results of each category in a single Ping point cloud data. First, obtain the various categories obtained by cluster analysis in the single Ping point cloud data, which represent the distribution areas of different targets or backgrounds in the data. Then, further calculate the three-dimensional hollow point coordinates of these categories, which provide the geometric center of the category space distribution. Finally, select and retain the most representative center point coordinates from these hollow point coordinates. This step not only simplifies the data, but also enhances the description of the core position of each category.
[0058] As a preferred embodiment of the second aspect, after performing a clustering operation on the dimension-reduced full Ping point cloud data, the method further includes:
[0059] Get the clustering results of each category in a single Ping point cloud data;
[0060] Calculate the intensity variance value of the clustering results of each category in the single Ping point cloud data;
[0061] Compare the intensity variance values of the clustering results of each category in the single Ping point cloud data;
[0062] According to the intensity variance value of the clustering results of each category in the single Ping point cloud data, the second underwater background and the second underwater moving target are separated, and the second underwater background is used to assist in judging the first underwater background; among them, the second underwater background has the largest intensity variance value.
[0063] As a preferred embodiment of the second aspect, the second underwater background is used to assist in determining the first underwater background, specifically:
[0064] The second underwater background is used as a strongly correlated confidence indicator for distinguishing the first underwater background.
[0065] In this preferred embodiment, the present application can identify different categories in a single Ping point cloud data by first performing a clustering operation on the dimensionally reduced full Ping point cloud data, which lays the foundation for further analysis. Subsequently, the intensity variance value of each clustering result is calculated. This step reveals the distribution consistency of the point cloud intensity within each category. By comparing the sizes of these variance values, it can be inferred that the cluster with a larger variance value may represent the underwater background, because the background area usually contains a variety of different reflection sources, while the cluster with a smaller variance value may represent a moving target in the water, because the reflection characteristics of the moving target are relatively consistent. Based on this reasoning, the cluster with the largest intensity variance value is identified as the second underwater background, and the cluster with the smallest intensity variance value is identified as the second underwater moving target. This method not only enhances the understanding of the underwater environment, but also provides an auxiliary means to verify and assist in judging the classification results of the underwater background and moving targets previously obtained by full Ping clustering, thereby improving the accuracy and reliability of the overall recognition process.
[0066] As a preferred embodiment of the second aspect, clustering operation is performed on the center point coordinates of each category in the single Ping point cloud data to obtain clustering results of each category of the full Ping point cloud data; according to the clustering results of each category of the full Ping point cloud data, the first underwater background and the first underwater moving target in the underwater initial full Ping point cloud data are determined, specifically:
[0067] Get the center point coordinates of each category in a single Ping point cloud data;
[0068] Display the center point coordinates of each category in all single Ping point cloud data in the same coordinate system to obtain a second coordinate system;
[0069] Initialize the domain radius value and domain point value of the DBSCAN clustering algorithm;
[0070] Using the DBSCAN clustering algorithm to cluster the center point coordinates of each category in the single Ping point cloud data in the second coordinate system, to obtain clustering results of each category of the full Ping point cloud data;
[0071] According to the clustering results of each category of the full Ping point cloud data, the center point coordinate change trend of the clustering results of each category of the full Ping point cloud data is analyzed, and the first underwater background and the first underwater moving target are separated according to the center point coordinate change trend; among them, the first underwater background has the smallest center point coordinate change trend, and the first underwater moving target has the largest center point coordinate change trend.
[0072] In this preferred embodiment, the present application effectively integrates the information in the full Ping point cloud data by accurately clustering the center point coordinates of each category in the single Ping point cloud data, thereby identifying the underwater background and underwater moving targets. First, the center point coordinates of each category in the single Ping are obtained, and these coordinates represent the spatial concentration trend of each cluster. Then, by initializing and applying the DBSCAN clustering algorithm, these center point coordinates are clustered within the full Ping data range, and the clustering results of each category of the full Ping point cloud data are obtained. Then, the change trend of the center point coordinates in these clustering results is analyzed. This step reveals the dynamic characteristics of the point cloud data in the time series. Finally, according to the change trend of the center point coordinates, the cluster with the smallest change trend is identified as the first underwater background, and the cluster with the largest change trend is identified as the first underwater moving target. This method utilizes the spatial and temporal information of the point cloud data, and through clustering and trend analysis, it improves the recognition accuracy of underwater static background and dynamic targets, and provides a more accurate and reliable technical means for monitoring and analyzing underwater environments. BRIEF DESCRIPTION OF THE DRAWINGS
[0073] Figure 1 : A schematic flow chart of an embodiment of a method for processing underwater moving targets provided by the present invention;
[0074] Figure 2 : A schematic diagram of the structure of the full Ping point cloud signal provided by the present invention;
[0075] Figure 3 : A schematic diagram of the structure of the full Ping point cloud signal after voxel downsampling provided by the present invention;
[0076] Figure 4 : A schematic diagram of the structure after XZ projection provided by the present invention;
[0077] Figure 5 : A schematic diagram of the structure of the clustered projection image provided by the present invention;
[0078] Figure 6 : A schematic diagram of the structure of the clustering result after denoising provided by the present invention;
[0079] Figure 7 : A schematic diagram of the process of the method for processing three-dimensional underwater moving targets provided by the present invention;
[0080] Figure 8 : A schematic structural diagram of an embodiment of the underwater moving target processing device provided by the present invention. DETAILED DESCRIPTION
[0081] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0082] Embodiment 1
[0083] Please refer to Figure 1 , which is a method for processing underwater moving targets provided by an embodiment of the present invention.
[0084] In this embodiment, the process of the underwater moving target processing method in this application is described in detail through steps S01-S03.
[0085] S01: Obtain the initial full-level point cloud data of the bottom. The obtained initial full-level point cloud data of the bottom is as follows: Figure 2 As shown;
[0086] S02: Preprocessing the initial full Ping point cloud data of the bottom to obtain dimension-reduced full Ping point cloud data;
[0087] More specifically, the preprocessing process can be expressed as follows:
[0088] Get the initial full ping point data of the bottom;
[0089] The initial full Ping point cloud data is subjected to voxel downsampling operation to obtain the downsampled full Ping point cloud data, such as Figure 3 As shown, the specific process of downsampling is: taking a cube with a grid size of 0.2 as the basic voxel unit, voxel downsampling operation is performed on X, Y, and Z in the full Ping point cloud data, each voxel unit retains at most one point, and when there are multiple data points inside the voxel unit, only the center point of these data points in space is selected. At the same time, if there is only one point data in the voxel unit, it is selected through the threshold judgment. The voxelization operation of the full Ping point cloud data can effectively realize the downsampling of data, and downsample hundreds of thousands or millions of point cloud data to less than 10,000, to prevent memory errors.
[0090] The downsampled full Ping point cloud data is projected in the XZ direction to obtain the reduced-dimensional full Ping point cloud data, such as Figure 4 As shown, the specific process of projection in the XZ direction is: projecting the full Ping point cloud data on the XZ plane, that is, directly discarding the Y component of the point cloud X, Y, and Z data, extracting the X and Z components, and distributing the projected points on the two-dimensional image, which can effectively reduce the amount of calculation.
[0091] This application first obtains the initial full Ping point cloud data of the bottom of the water, and effectively reduces the amount of data through voxelized downsampling operations. This step not only reduces the complexity of the data, but also helps to remove noise and retain key spatial features. Subsequently, the downsampled point cloud data is projected in the XZ direction to further convert the three-dimensional data into a two-dimensional representation. This dimensionality reduction process not only simplifies the data processing requirements, but also makes the visualization and subsequent analysis of the data more intuitive and efficient. This preprocessing process provides a structured and easy-to-operate format for the point cloud data of the underwater environment, laying a solid foundation for the accurate identification and analysis of the underwater background and moving targets.
[0092] S03: performing a clustering operation on the dimension-reduced full Ping point cloud data to obtain the coordinates of the cluster center points of each category in the single Ping point cloud data;
[0093] Perform clustering operation on the coordinates of cluster center points of each category in single Ping point cloud data to obtain clustering results of each category of the whole Ping point cloud data;
[0094] According to the clustering results of each category of the full Ping point cloud data, the first underwater background and the first underwater moving target in the initial full Ping point cloud data are determined.
[0095] Furthermore, the clustering operation is performed on the dimension-reduced full Ping point cloud data to obtain the cluster center point coordinates of each category in the single Ping point cloud data, specifically:
[0096] The reduced-dimensional full Ping point cloud data is clustered. Specifically, after directional projection of the full Ping point cloud data, DBSCAN (density-based clustering algorithm) is used to cluster the points on the two-dimensional image. The field radius eps is selected as 2, and the number of points in the field min_sample is selected as 5. The point cloud data projected into the image is clustered into several categories, and the clustering results of each category in the single Ping point cloud data are obtained, such as Figure 5 By inferring the underwater bottom, underwater target and underwater noise signals under each single Ping signal, the underwater noise signals can be further suppressed, providing a basis for the subsequent underwater target signal processing.
[0097] Calculate the three-dimensional center point coordinates of the clustering results of each category in the single Ping point cloud data respectively;
[0098] The three-dimensional center point coordinates are uniformly displayed in one coordinate system to obtain a first coordinate system;
[0099] The coordinates of all Ping center points are clustered in the first coordinate system to obtain the coordinates of cluster center points of each category in the single Ping point cloud data.
[0100] This application can accurately locate the central trend of each category in three-dimensional space by calculating the center point coordinates of the clustering results of each category in a single Ping point cloud data. First, obtain the various categories obtained by clustering analysis in the single Ping point cloud data, which represents the distribution areas of different targets or backgrounds in the data. Then, further calculate the three-dimensional hollow point coordinates of these categories, which provide the geometric center of the category space distribution. Finally, select and retain the most representative center point coordinates from these hollow point coordinates. This step not only simplifies the data, but also enhances the description of the core position of each category.
[0101] Furthermore, after performing a clustering operation on the dimension-reduced full Ping point cloud data, it also includes:
[0102] Get the clustering results of each category in a single Ping point cloud data;
[0103] Calculate the intensity variance value of the clustering results of each category in the single Ping point cloud data;
[0104] Compare the intensity variance values of the clustering results of each category in the single Ping point cloud data;
[0105] According to the intensity variance value of the clustering results of each category in the single Ping point cloud data, the second underwater background and the second underwater moving target are separated, and the second underwater background is used to assist in judging the first underwater background; among them, the second underwater background has the largest intensity variance value.
[0106] Further, the second underwater background is used to assist in determining the first underwater background, specifically:
[0107] The second underwater background is used as a strong correlation confidence indicator for the first underwater background. The second underwater background is calculated using the variance of the intensity value, which provides strong confidence support for the determination of the first underwater background. When an area is confirmed as the second underwater background, this significantly increases the possibility that the area also belongs to the first underwater background. Therefore, the recognition result of the second underwater background not only enhances the confidence in the classification of the first underwater background, but also serves as a reliable indicator to assist the confirmation and decision-making process.
[0108] If the second underwater background is inconsistent with the first underwater background, the accuracy of the original judgment can be verified by recalculating the first underwater background, ensuring the accuracy of the recognition process and eliminating possible calculation errors. This review mechanism ensures the reliability and effectiveness of the final judgment.
[0109] This application can identify different categories in a single Ping point cloud data by first performing a clustering operation on the reduced-dimensional full Ping point cloud data, which lays the foundation for further analysis. Subsequently, the intensity variance value of each clustering result is calculated. This step reveals the distribution consistency of the point cloud intensity within each category. By comparing the sizes of these variance values, it can be inferred that clusters with larger variance values may represent underwater backgrounds, because the background area usually contains a variety of different reflection sources, while clusters with smaller variance values may represent moving targets in the water, because the reflection characteristics of moving targets are relatively consistent. Based on this reasoning, the cluster with the largest intensity variance value is identified as the second underwater background, and the cluster with the smallest intensity variance value is identified as the second underwater moving target. This method not only enhances the understanding of the underwater environment, but also provides an auxiliary means to verify and assist in judging the classification results of the underwater background and moving targets previously obtained through full Ping clustering, thereby improving the accuracy and reliability of the overall recognition process.
[0110] Furthermore, the center point coordinates of each category in the single Ping point cloud data are clustered to obtain clustering results of each category of the full Ping point cloud data; according to the clustering results of each category of the full Ping point cloud data, the first underwater background and the first underwater moving target in the underwater initial full Ping point cloud data are determined, specifically:
[0111] Get the center point coordinates of each category in a single Ping point cloud data;
[0112] Display the center point coordinates of each category in all single Ping point cloud data in the same coordinate system to obtain a second coordinate system;
[0113] Initialize the DBSCAN clustering algorithm with a domain radius value of 2 and a domain point value of 3;
[0114] The DBSCAN clustering algorithm is used to cluster the center point coordinates of each category in the single Ping point cloud data in the second coordinate system to obtain the clustering results of each category of the full Ping point cloud data, such as Figure 6 As shown, the blue is the underwater background after the center points of all pings are clustered, and the orange is the underwater moving target;
[0115] According to the clustering results of each category of the full Ping point cloud data, the association relationship between Ping classes is established, and the center point coordinate change trend of the clustering results of each category of the full Ping point cloud data is analyzed. The first underwater background and the first underwater moving target are separated according to the center point coordinate change trend; among them, the first underwater background has the smallest center point coordinate change trend, and the first underwater moving target has the largest center point coordinate change trend.
[0116] This application effectively integrates the information in the full Ping point cloud data by accurately clustering the center point coordinates of each category in the single Ping point cloud data, thereby identifying the underwater background and underwater moving targets. First, the center point coordinates of each category in the single Ping are obtained, and these coordinates represent the spatial concentration trend of each cluster. Then, by initializing and applying the DBSCAN clustering algorithm, these center point coordinates are clustered within the full Ping data range, and the clustering results of each category of the full Ping point cloud data are obtained. Then, the changing trend of the center point coordinates in these clustering results is analyzed. This step reveals the dynamic characteristics of the point cloud data in the time series. Finally, according to the changing trend of the center point coordinates, the cluster with the smallest changing trend is identified as the first underwater background, and the cluster with the largest changing trend is identified as the first underwater moving target. This method utilizes the spatial and temporal information of the point cloud data, and through clustering and trend analysis, it improves the recognition accuracy of underwater static background and dynamic targets, and provides a more accurate and reliable technical means for monitoring and analyzing underwater environments.
[0117] Furthermore, the overall process diagram of underwater moving target processing is as follows: Figure 7 As shown in the figure; first, the data density is reduced by voxel downsampling technology, and then the three-dimensional point cloud data is projected onto a two-dimensional plane to simplify subsequent processing. The DBSCAN algorithm is used to cluster the two-dimensional image to identify different categories such as the bottom, underwater targets, and noise. Subsequently, the intensity variance and center point coordinates are calculated independently for each Ping data, where the intensity variance value is used to assist in judging the accuracy of the final separation result; and by reapplying the DBSCAN clustering algorithm to the center point coordinates of all Pings, the correlation across Pings is established. Finally, by fusing the analysis results of a single Ping data, the moving targets in the water are effectively separated and identified, while the background and noise are distinguished, providing an efficient preprocessing strategy for underwater monitoring and target tracking.
[0118] This method first achieves data dimensionality reduction through a preprocessing step, converting complex three-dimensional point cloud data into a two-dimensional form that is easier to process, thereby reducing the complexity of the calculation and retaining key spatial information. Next, a clustering operation is performed on the full Ping point cloud data after dimensionality reduction to effectively divide the data into different categories, laying the foundation for subsequent analysis. The center point coordinates of each category are further calculated to provide a representative location reference for the point cloud data within each single Ping. By clustering these center point coordinates again, this method can identify the spatial distribution characteristics of the underwater background and moving targets on the scale of the full Ping data. Finally, combined with the full Ping clustering results, the present application can distinguish and determine the first underwater background and the first underwater moving target in the initial full Ping point cloud data of the underwater. This process not only improves the accuracy of underwater target detection, but also solves the problem of difficulty in separating three-dimensional underwater moving targets in the prior art.
[0119] Embodiment 2
[0120] Please refer to Figure 8 , which is a structural schematic diagram of an embodiment of the underwater moving target processing device provided in this application.
[0121] In this embodiment, the underwater moving target processing device includes an acquisition module 10 , a pre-processing module 20 and a calculation module 30 .
[0122] The acquisition module 10 is used to acquire the initial full-level point cloud data of the bottom. The acquired initial full-level point cloud data of the bottom is as follows: Figure 2 As shown;
[0123] The preprocessing module 20 is used to preprocess the initial full Ping point cloud data of the bottom to obtain the reduced-dimensional full Ping point cloud data;
[0124] More specifically, the preprocessing process can be expressed as follows:
[0125] Get the initial full ping point data of the bottom;
[0126] The initial full Ping point cloud data is subjected to voxel downsampling operation to obtain the downsampled full Ping point cloud data, such as Figure 3As shown, the specific process of downsampling is: taking a cube with a grid size of 0.2 as the basic voxel unit, voxel downsampling operation is performed on X, Y, and Z in the full Ping point cloud data, and each voxel unit retains at most one point. When there are multiple data points inside the voxel unit, only the center point of these data points in space is selected. At the same time, if there is only one point data in the voxel unit through the threshold, the data point is abandoned. The voxelization operation of the full Ping point cloud data can effectively realize data downsampling, and hundreds of thousands or millions of point cloud data can be downsampled to less than 10,000 to prevent memory errors.
[0127] The downsampled full Ping point cloud data is projected in the XZ direction to obtain the reduced-dimensional full Ping point cloud data, such as Figure 4 As shown, the specific process of projection in the XZ direction is: projecting the full Ping point cloud data on the XZ plane, that is, directly discarding the Y component of the point cloud X, Y, and Z data, extracting the X and Z components, and distributing the projected points on the two-dimensional image, which can effectively reduce the amount of calculation.
[0128] This application first obtains the initial full Ping point cloud data of the bottom of the water, and effectively reduces the amount of data through voxelized downsampling operations. This step not only reduces the complexity of the data, but also helps to remove noise and retain key spatial features. Subsequently, the downsampled point cloud data is projected in the XZ direction to further convert the three-dimensional data into a two-dimensional representation. This dimensionality reduction process not only simplifies the data processing requirements, but also makes the visualization and subsequent analysis of the data more intuitive and efficient. This preprocessing process provides a structured and easy-to-operate format for the point cloud data of the underwater environment, laying a solid foundation for the accurate identification and analysis of the underwater background and moving targets.
[0129] The calculation module 30 is used to perform a clustering operation on the reduced-dimensional full Ping point cloud data to obtain the coordinates of the cluster center points of each category in the single Ping point cloud data;
[0130] Perform clustering operation on the coordinates of cluster center points of each category in single Ping point cloud data to obtain clustering results of each category of the whole Ping point cloud data;
[0131] According to the clustering results of each category of the full Ping point cloud data, the first underwater background and the first underwater moving target in the initial full Ping point cloud data are determined.
[0132] Furthermore, the clustering operation is performed on the dimension-reduced full Ping point cloud data to obtain the cluster center point coordinates of each category in the single Ping point cloud data, specifically:
[0133] The reduced-dimensional full Ping point cloud data is clustered. Specifically, after directional projection of the full Ping point cloud data, DBSCAN (density-based clustering algorithm) is used to cluster the points on the two-dimensional image. The field radius eps is selected as 2, and the number of points in the field min_sample is selected as 5. The point cloud data projected into the image is clustered into several categories, and the clustering results of each category in the single Ping point cloud data are obtained, such as Figure 5 By inferring the underwater bottom, underwater target and underwater noise signals under each single Ping signal, the underwater noise signals can be further suppressed, providing a basis for the subsequent underwater target signal processing.
[0134] Calculate the three-dimensional center point coordinates of the clustering results of each category in the single Ping point cloud data respectively;
[0135] The three-dimensional center point coordinates are uniformly displayed in one coordinate system to obtain a first coordinate system;
[0136] The coordinates of all Ping center points are clustered in the first coordinate system to obtain the coordinates of cluster center points of each category in the single Ping point cloud data.
[0137] PingThis application can accurately locate the central trend of each category in three-dimensional space by calculating the center point coordinates of the clustering results of each category in a single Ping point cloud data. First, obtain the various categories obtained by clustering analysis in the single Ping point cloud data, which represents the distribution areas of different targets or backgrounds in the data. Then, further calculate the three-dimensional hollow point coordinates of these categories, which provide the geometric center of the category space distribution. Finally, select and retain the most representative center point coordinates from these hollow point coordinates. This step not only simplifies the data, but also enhances the description of the core position of each category.
[0138] Furthermore, after performing a clustering operation on the dimension-reduced full Ping point cloud data, it also includes:
[0139] Get the clustering results of each category in a single Ping point cloud data;
[0140] Calculate the intensity variance value of the clustering results of each category in the single Ping point cloud data;
[0141] Compare the intensity variance values of the clustering results of each category in the single Ping point cloud data;
[0142] According to the intensity variance value of the clustering results of each category in the single Ping point cloud data, the second underwater background and the second underwater moving target are separated, and the second underwater background is used to assist in judging the first underwater background; among them, the second underwater background has the largest intensity variance value.
[0143] Further, the second underwater background is used to assist in determining the first underwater background, specifically:
[0144] The second underwater background is used as a strong correlation confidence indicator for the first underwater background. The second underwater background is calculated using the variance of the intensity value, which provides strong confidence support for the determination of the first underwater background. When an area is confirmed as the second underwater background, this significantly increases the possibility that the area also belongs to the first underwater background. Therefore, the recognition result of the second underwater background not only enhances the confidence in the classification of the first underwater background, but also serves as a reliable indicator to assist the confirmation and decision-making process.
[0145] If the second underwater background is inconsistent with the first underwater background, the accuracy of the original judgment can be verified by recalculating the first underwater background, ensuring the accuracy of the recognition process and eliminating possible calculation errors. This review mechanism ensures the reliability and effectiveness of the final judgment.
[0146] This application can identify different categories in a single Ping point cloud data by first performing a clustering operation on the reduced-dimensional full Ping point cloud data, which lays the foundation for further analysis. Subsequently, the intensity variance value of each clustering result is calculated. This step reveals the distribution consistency of the point cloud intensity within each category. By comparing the sizes of these variance values, it can be inferred that clusters with larger variance values may represent underwater backgrounds, because the background area usually contains a variety of different reflection sources, while clusters with smaller variance values may represent moving targets in the water, because the reflection characteristics of moving targets are relatively consistent. Based on this reasoning, the cluster with the largest intensity variance value is identified as the second underwater background, and the cluster with the smallest intensity variance value is identified as the second underwater moving target. This method not only enhances the understanding of the underwater environment, but also provides an auxiliary means to verify and assist in judging the classification results of the underwater background and moving targets previously obtained through full Ping clustering, thereby improving the accuracy and reliability of the overall recognition process.
[0147] Furthermore, the center point coordinates of each category in the single Ping point cloud data are clustered to obtain clustering results of each category of the full Ping point cloud data; according to the clustering results of each category of the full Ping point cloud data, the first underwater background and the first underwater moving target in the underwater initial full Ping point cloud data are determined, specifically:
[0148] Get the center point coordinates of each category in a single Ping point cloud data;
[0149] Display the center point coordinates of each category in all single Ping point cloud data in the same coordinate system to obtain a second coordinate system;
[0150] Initialize the DBSCAN clustering algorithm with a domain radius value of 2 and a domain point value of 3;
[0151] The DBSCAN clustering algorithm is used to cluster the center point coordinates of each category in the single Ping point cloud data in the second coordinate system to obtain the clustering results of each category of the full Ping point cloud data, such as Figure 6 As shown, the blue is the underwater background after the center points of all pings are clustered, and the orange is the underwater moving target;
[0152] According to the clustering results of each category of the full Ping point cloud data, the association relationship between Ping classes is established, and the center point coordinate change trend of the clustering results of each category of the full Ping point cloud data is analyzed. The first underwater background and the first underwater moving target are separated according to the center point coordinate change trend; among them, the first underwater background has the smallest center point coordinate change trend, and the first underwater moving target has the largest center point coordinate change trend.
[0153] This application effectively integrates the information in the full Ping point cloud data by accurately clustering the center point coordinates of each category in the single Ping point cloud data, thereby identifying the underwater background and underwater moving targets. First, the center point coordinates of each category in the single Ping are obtained, and these coordinates represent the spatial concentration trend of each cluster. Then, by initializing and applying the DBSCAN clustering algorithm, these center point coordinates are clustered within the full Ping data range, and the clustering results of each category of the full Ping point cloud data are obtained. Then, the changing trend of the center point coordinates in these clustering results is analyzed. This step reveals the dynamic characteristics of the point cloud data in the time series. Finally, according to the changing trend of the center point coordinates, the cluster with the smallest changing trend is identified as the first underwater background, and the cluster with the largest changing trend is identified as the first underwater moving target. This method utilizes the spatial and temporal information of the point cloud data, and through clustering and trend analysis, it improves the recognition accuracy of underwater static background and dynamic targets, and provides a more accurate and reliable technical means for monitoring and analyzing underwater environments.
[0154] Furthermore, the overall process diagram of underwater moving target processing is as follows: Figure 7As shown in the figure; first, the data density is reduced by voxel downsampling technology, and then the three-dimensional point cloud data is projected onto a two-dimensional plane to simplify subsequent processing. The DBSCAN algorithm is used to cluster the two-dimensional image to identify different categories such as the bottom, underwater targets, and noise. Subsequently, the intensity variance and center point coordinates are calculated independently for each Ping data, where the intensity variance value is used to assist in judging the accuracy of the final separation result; and by reapplying the DBSCAN clustering algorithm to the center point coordinates of all Pings, the correlation across Pings is established. Finally, by fusing the analysis results of a single Ping data, the moving targets in the water are effectively separated and identified, while the background and noise are distinguished, providing an efficient preprocessing strategy for underwater monitoring and target tracking.
[0155] This method first achieves data dimensionality reduction through a preprocessing step, converting complex three-dimensional point cloud data into a two-dimensional form that is easier to process, thereby reducing the complexity of the calculation and retaining key spatial information. Next, a clustering operation is performed on the full Ping point cloud data after dimensionality reduction to effectively divide the data into different categories, laying the foundation for subsequent analysis. The center point coordinates of each category are further calculated to provide a representative location reference for the point cloud data within each single Ping. By clustering these center point coordinates again, this method can identify the spatial distribution characteristics of the underwater background and moving targets on the scale of the full Ping data. Finally, combined with the full Ping clustering results, the present application can distinguish and determine the first underwater background and the first underwater moving target in the initial full Ping point cloud data of the underwater. This process not only improves the accuracy of underwater target detection, but also solves the problem of difficulty in separating three-dimensional underwater moving targets in the prior art.
[0156] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.
Claims
1. A method for processing underwater moving targets, characterized in that: include: Get the initial full Ping point cloud data of the bottom; Preprocess the initial full Ping point cloud data of the bottom to obtain the reduced-dimensional full Ping point cloud data; Performing a clustering operation on the dimension-reduced full Ping point cloud data to obtain the coordinates of the cluster center points of each category in the single Ping point cloud data; Perform clustering operation on the coordinates of cluster center points of each category in single Ping point cloud data to obtain clustering results of each category of the whole Ping point cloud data; According to the clustering results of each category of the full Ping point cloud data, the first underwater background and the first underwater moving target in the initial full Ping point cloud data are determined.
2. The underwater moving target processing method according to claim 1, characterized in that: The preprocessing of the initial full Ping point cloud data of the bottom to obtain the reduced-dimensional full Ping point cloud data is specifically as follows: Get the initial full ping point data of the bottom; Perform voxel downsampling operation on the initial full Ping point cloud data to obtain downsampled full Ping point cloud data; The downsampled full Ping point cloud data is projected in the XZ direction to obtain the reduced-dimensional full Ping point cloud data.
3. The underwater moving target processing method according to claim 1, characterized in that: The clustering operation is performed on the dimension-reduced full Ping point cloud data to obtain the cluster center point coordinates of each category in the single Ping point cloud data, specifically: Performing a clustering operation on the dimension-reduced full Ping point cloud data to obtain clustering results of each category in the single Ping point cloud data; Calculate the three-dimensional center point coordinates of the clustering results of each category in the single Ping point cloud data respectively; The three-dimensional center point coordinates are uniformly displayed in one coordinate system to obtain a first coordinate system; The coordinates of all Ping center points are clustered in the first coordinate system to obtain the coordinates of cluster center points of each category in the single Ping point cloud data.
4. The underwater moving target processing method according to claim 1, characterized in that: After clustering the dimension-reduced full Ping point cloud data, the method further includes: Get the clustering results of each category in a single Ping point cloud data; Calculate the intensity variance value of the clustering results of each category in the single Ping point cloud data; Compare the intensity variance values of the clustering results of each category in the single Ping point cloud data; According to the intensity variance value of the clustering results of each category in the single Ping point cloud data, the second underwater background and the second underwater moving target are separated, and the second underwater background is used to assist in judging the first underwater background; among them, the second underwater background has the largest intensity variance value.
5. The underwater moving target processing method according to claim 4 is characterized in that: The second underwater background is used to assist in determining the first underwater background, specifically: The second underwater background is used as a strongly correlated confidence indicator for distinguishing the first underwater background.
6. The underwater moving target processing method according to claim 1, 4 or 5, characterized in that: The clustering operation is performed on the center point coordinates of each category in the single Ping point cloud data to obtain the clustering results of each category of the full Ping point cloud data; According to the clustering results of each category of the full Ping point cloud data, the first underwater background and the first underwater moving target in the initial full Ping point cloud data are determined, specifically: Get the center point coordinates of each category in a single Ping point cloud data; Display the center point coordinates of each category in all single Ping point cloud data in the same coordinate system to obtain a second coordinate system; Initialize the domain radius value and domain point value of the DBSCAN clustering algorithm; Using the DBSCAN clustering algorithm to cluster the center point coordinates of each category in the full Ping point cloud data in the second coordinate system, to obtain clustering results of each category of the full Ping point cloud data; According to the clustering results of each category of the full Ping point cloud data, the center point coordinate change trend of the clustering results of each category of the full Ping point cloud data is analyzed, and the first underwater background and the first underwater moving target are separated according to the center point coordinate change trend; among them, the first underwater background has the smallest center point coordinate change trend, and the first underwater moving target has the largest center point coordinate change trend.
7. An underwater moving target processing device, characterized in that: It includes an acquisition module, a preprocessing module and a calculation module; The acquisition module is used to obtain the initial full Ping point cloud data of the bottom; The preprocessing module is used to preprocess the initial full Ping point cloud data of the bottom to obtain the reduced-dimensional full Ping point cloud data; The calculation module is used to perform clustering operation on the dimension-reduced full Ping point cloud data to obtain the coordinates of the cluster center points of each category in the single Ping point cloud data; Perform clustering operation on the coordinates of cluster center points of each category in single Ping point cloud data to obtain clustering results of each category of the whole Ping point cloud data; According to the clustering results of each category of the full Ping point cloud data, the first underwater background and the first underwater moving target in the initial full Ping point cloud data are determined.
8. The underwater moving target processing device according to claim 7, characterized in that: The preprocessing of the initial full Ping point cloud data of the bottom to obtain the reduced-dimensional full Ping point cloud data is specifically as follows: Get the initial full ping point data of the bottom; Perform voxel downsampling operation on the initial full Ping point cloud data to obtain downsampled full Ping point cloud data; The downsampled full Ping point cloud data is projected in the XZ direction to obtain the reduced-dimensional full Ping point cloud data.
9. The underwater moving target processing device according to claim 7, characterized in that: The clustering operation is performed on the dimension-reduced full Ping point cloud data to obtain the cluster center point coordinates of each category in the single Ping point cloud data, specifically: Performing a clustering operation on the dimension-reduced full Ping point cloud data to obtain clustering results of each category in the single Ping point cloud data; Calculate the three-dimensional center point coordinates of the clustering results of each category in the single Ping point cloud data respectively; The three-dimensional center point coordinates are uniformly displayed in one coordinate system to obtain a first coordinate system; The coordinates of all Ping center points are clustered in the first coordinate system to obtain the coordinates of cluster center points of each category in the single Ping point cloud data.
10. The underwater moving target processing device according to claim 7, characterized in that: After clustering the dimension-reduced full Ping point cloud data, the method further includes: Get the clustering results of each category in a single Ping point cloud data; Calculate the intensity variance value of the clustering results of each category in the single Ping point cloud data; Compare the intensity variance values of the clustering results of each category in the single Ping point cloud data; According to the intensity variance value of the clustering results of each category in the single Ping point cloud data, the second underwater background and the second underwater moving target are separated, and the second underwater background is used to assist in judging the first underwater background; among them, the second underwater background has the largest intensity variance value.
11. The underwater moving target processing device according to claim 10, characterized in that: The second underwater background is used to assist in determining the first underwater background, specifically: The second underwater background is used as a strongly correlated confidence indicator for distinguishing the first underwater background.
12. The underwater moving target processing device according to claim 7, 10 or 11, characterized in that: The clustering operation is performed on the center point coordinates of each category in the single Ping point cloud data to obtain the clustering results of each category of the full Ping point cloud data; According to the clustering results of each category of the full Ping point cloud data, the first underwater background and the first underwater moving target in the initial full Ping point cloud data are determined, specifically: Get the center point coordinates of each category in a single Ping point cloud data; Display the center point coordinates of each category in all single Ping point cloud data in the same coordinate system to obtain a second coordinate system; Initialize the domain radius value and domain point value of the DBSCAN clustering algorithm; Using the DBSCAN clustering algorithm to cluster the center point coordinates of each category in the full Ping point cloud data in the second coordinate system, to obtain clustering results of each category of the full Ping point cloud data; According to the clustering results of each category of the full Ping point cloud data, the center point coordinate change trend of the clustering results of each category of the full Ping point cloud data is analyzed, and the first underwater background and the first underwater moving target are separated according to the center point coordinate change trend; among them, the first underwater background has the smallest center point coordinate change trend, and the first underwater moving target has the largest center point coordinate change trend.
Citation Information
Patent Citations
ViBe-based three-dimensional sonar point cloud image segmentation method
CN108537814A
Obstacle detection method and system
CN110458055A
Underwater close-range target recognition and tracking method and system based on deep fusion
CN113792593A
Target detection model optimization method and device, electronic equipment and storage medium
CN115527051A
Millimeter wave radar point cloud clustering and multi-target recognition machine learning method
CN116012628A
